Home Theater Forum and Systems banner

1 - 20 of 116 Posts

·
Premium Member
Joined
·
129 Posts
Discussion Starter #1
Dear John:

Many thanks for your continued advancements in performance in REW! If I understand correctly reading the latest beta notes, you have conquered the OSX measurement issues with Java that plagued us before and led me to use ASIO on PC?

--------------

I'd like to make my "annual" request to see if REW can please implement frequency dependent windowing. Attached is a slide from Dr. Uli Brueggemann showing his example.

15 cycles (at 48 kHz I believe) is Uli's favorite setting. It provides sufficient psychoacoustic width for low frequencies and sufficiently "anechoic" width for the ear at high frequencies.

I'm usually using a 500 ms right hand window in REW which is probably psychoacoustically accurate for low frequency measurements. But the high frequency display is too smooth. So I tend to ignore or discount the high frequency measurements as they are not "resolved" enough with such a large window. It would be nice to see it all in one graph.
 

Attachments

·
Premium Member
Joined
·
129 Posts
Discussion Starter #3
I took a good look at the variable smoothing. My standard of excellence for sonic performance and psychoacoustic accuracy continues to be Acourate and Acourate Convolver. You might ask "so why are you continuing to use Room EQ Wizard?" The answers are simply: To cross check, and because REW offers a more effective distortion measurement, wonderful waterfall displays and is far easier to use than Acourate!

Also, I really don't know enough about the differences between smoothing and windowing to say with assurance that variable smoothing is a compromise, but it seems that way to me because the ear responds to the direct sound of the loudspeaker above the room's nominal Schroeder frequency and at 20 kHz Jim Johnston says to use a very short window, near 1 ms if possible. If you start with a 500 ms window but you smooth the result is that the same or effectively the same as using a 1 ms window? I don't think so, but I don't have the math to say for sure.


So I tried the new variable smoothing and my conclusions are (based on listening and visual comparison), that variable smoothing is not smoothed enough below 1 kHz, and oversmoothed above 1 kHz. I'm using a 500 ms right hand Hann window in REW. Attached are three images, all of the post-corrected front left speaker, corrected by Acourate.

Attached:

The first is an image of Acourate's psychoacoustic amplitude response display, using 15 samples at 1 kHz for the variable window. I exported a 2448 impulse from REW, loaded it into Acourate. First I then had to do a cut n' window because the impulse was far too long for Acourate to display a frequency response. I windowed it very very wide, so none of the actual information was cut off.

The second is REW's display of the same information, variable window.

The third is with 1/6 octave smoothing. 1/6 octave appears to be closer to Acourate's psychoacoustic measurement. Further investigation is needed, because this is of an impulse-corrected loudspeaker running through a convolver, so it has been considerably "smoothed" before REW could even measure it.
 

Attachments

·
Premium Member
Joined
·
6,711 Posts
If I understand correctly reading the latest beta notes, you have conquered the OSX measurement issues with Java that plagued us before and led me to use ASIO on PC?
That's right, OS X is now well behaved thanks to Oracle's Java runtime, which is bundled inside the REW download so Java does not need to be installed.

I'm usually using a 500 ms right hand window in REW which is probably psychoacoustically accurate for low frequency measurements. But the high frequency display is too smooth.
That's a little odd, since wide windows usually give ragged HF responses due to the comb filtering effects of reflections that fall within the window.

Variable windowing on its own isn't enough to give a psycho-acoustic response, other factors need to be taken into account such as removing or reducing narrow dips, which remain present in variable windowing (per images below - complex smoothing is mathematically equivalent to variable windowing). Variable windowing and variable smoothing are not equivalent, but smoothing provides better EQ targets from the tests I have done. In REW the variable smoothing uses no smoothing below 100 Hz, then varies from 1/48 octave at 100 Hz to 1/3 octave at and above 10 kHz. At 1 kHz the smoothing is 1/6 octave.



 

·
Premium Member
Joined
·
129 Posts
Discussion Starter #5
That's right, OS X is now well behaved thanks to Oracle's Java runtime, which is bundled inside the REW download so Java does not need to be installed.
That's fantastic, John. A great solution to a nagging problem. I will have to try sampling soon on OSX with REW!

That's a little odd, since wide windows usually give ragged HF responses due to the comb filtering effects of reflections that fall within the window.
Please take a look at the much-too-smooth response above 1 kHz in my image posted just a little above. Notice that even 1/6 octave shows more detail above 1 kHz than variable smoothing in its current incarnation.

Variable windowing on its own isn't enough to give a psycho-acoustic response, other factors need to be taken into account such as removing or reducing narrow dips, which remain present in variable windowing (per images below - complex smoothing is mathematically equivalent to variable windowing).
I'll take your word for it that complex smoothing is mathematically equivalent to variable windowing. So, is REW going to be able to do "complex smoothing"? The inaudible narrow dips are somehow taken care of by Uli Brueggemann. I don't know if it is his choice of window below 100 Hz or some other trick.... Regardless, it appears that with your current variable smoothing, set as you describe:

Variable windowing and variable smoothing are not equivalent, but smoothing provides better EQ targets from the tests I have done. In REW the variable smoothing uses no smoothing below 100 Hz, then varies from 1/48 octave at 100 Hz to 1/3 octave at and above 10 kHz. At 1 kHz the smoothing is 1/6 octave.
....is much too "aggressive" (compared to the perception) below 100 Hz and so needs some smoothing and much too smooth above 1 kHz, does not show as much detail as the ear perceives. I think that the irregularities of the direct response of a tweeter without room reflection issues are audible and of concern.

Regardless of whether you agree with Uli Brueggemann's particular "psychoacoustic" settings, from my point of view it would be nice to be able to set the parameters of REW's smoothing display so that it would be closer in display to Acourate's choices, if that is desired. Do you think that some parameters of complex smoothing could be set by the user and saved as a preset if desired? I do grant that at some point it becomes a completely subjective discussion. Nevertheless, I do have ears, and Acourate is the first system I have EVER encountered (in over 40 years of working with corrected and non-corrected audio systems) which corrects without requiring any further intervention on my part! That's a pretty remarkable endorsement.
 

·
Premium Member
Joined
·
129 Posts
Discussion Starter #6
Dear John: As much as I would like to love variable smoothing, it is not the same as a variable window. I conferred with Jim Johnston and basically he said that with the right parameters the results can be somewhat comparable, but the devil is in the details. I am not getting even close to Acourate's measurements at the high end. Basically I would like to marry a long FFT window length circa 200 to 500 ms below say, 1 kHz, with a short FFT window length circa 10 ms above 1 kHz.

Is there any way we can splice together two (or more) frequency responses? Or can you come up with a couple default pairs of windows? All the psychoacoustic papers by JJ and others make it clear that the ear responds to the earliest signals and not the room (near anechoic) at high frequencies, but the ear integrates the room at low frequencies. This means that a variable window will be far more psychoacoustically accurate. I hope you will be able to accomplish this sooner or later. Thanks!
 

·
Super Moderator
Joined
·
1,673 Posts
Dear John: As much as I would like to love variable smoothing, it is not the same as a variable window. I conferred with Jim Johnston and basically he said that with the right parameters the results can be somewhat comparable, but the devil is in the details. I am not getting even close to Acourate's measurements at the high end. Basically I would like to marry a long FFT window length circa 200 to 500 ms below say, 1 kHz, with a short FFT window length circa 10 ms above 1 kHz.

Is there any way we can splice together two (or more) frequency responses? Or can you come up with a couple default pairs of windows? All the psychoacoustic papers by JJ and others make it clear that the ear responds to the earliest signals and not the room (near anechoic) at high frequencies, but the ear integrates the room at low frequencies. This means that a variable window will be far more psychoacoustically accurate. I hope you will be able to accomplish this sooner or later. Thanks!
Hi Bob,
You've been talking to the right people.:)
Did JJ suggest 1k, or more around 700Hz? TIA

cheers
 

·
Registered
Joined
·
106 Posts
Bobkatz,

Hi, my name is Nick. I very often read this and other forums but rarely do I chime in on a topic. Most of the other people on this forum are more experienced with the general types of questions I read, and I learn a lot by just reading all the Q & A. However, your question of frequency dependent windowing caught my interest as this is something I have a daily working experience with. Have you ever heard of SysTune? It's software from AFMG, a Berlin based software company that makes all kinds of awesome software that's used in the loudspeaker design and installation industry. SysTune is primarily geared toward the pro sound industry, but, I use it. And I work as an installer of custom car audio systems. Frequency dependent windowing, in SysTune this is called the TFC, which stands for time frequency constant. It works like this, when you set the window there are 3 markers that represent the right hand window. One for 8 kHz, one for 1 kHz, and one for 125Hz. Now those are just markers to kinda give you an idea of the "spread" of the TFC. In actuality it is a constantly variable time window without discrete steps. Whatever the window time you set for 8 kHz, it is twice as long for half that freq. For example, say you set your 8 kHz marker for 2ms after the peak of the IR, then your window time for for 4 kHz will be 4ms, 2 kHz-8ms, 1 kHz-16ms, and so on. This way, you can window out reflections but still keep full range response in one graph. Well, window out higher freq. reflections anyway. So as you can imagine, in a car, the reflected energy can be REALLY close to the direct arrival. So with conventional windowing, in a car, if you windowed out the reflections, you'd end up with barely useable data, it would be a time consuming process to properly perform a phase alignment of mid to tweeter in a fully active system for example. And then have to change your settings to see the next driver's interactions as you worked down the freq. scale. But with TFC, it works brilliantly. Another one that is similar to SysTune is Smaart v7. It is a program that uses what's called Multi Time Window, or MTW. however, with Smaart, the multi time window is not user adjustable, so while I do use that software at work for certain things, SysTune is my main program. Room EQ Wizard plays an integral role in my work by allowing me to create compensation files for different mics and inputs, and also for its trace arithmetic, which I use to find the cabin gain of different vehicles. Anyway, thought just in case you hadn't heard of SysTune (or Smaart) that you might find them interesting, as the very mechanism they use to perform their measurements is variable time windowing!
 

·
Premium Member
Joined
·
129 Posts
Discussion Starter #9
Hi Bob,
You've been talking to the right people.:)
Did JJ suggest 1k, or more around 700Hz? TIA

cheers
All I remember from talking with JJ is a window width of up to 500 ms below about 200 Hz and very short (1 or 2 ms) at 20 kHz. Yes, the devil is in the details. Windowing is NOT the same as smoothing and in JJ's AES papers he makes that clear. If you are not looking at the earliest sound first it's not the same as smoothing the whole kit and kaboodle!

JJ also told me that with the right windowing that a subjectively flat response will correspond with a measured flat response. Perhaps someday he will enlighten us how to do that. However, I've never experienced that and not being the super mathematician that JJ is I just have learned to accept that I have to use a particular high frequency rolloff with a particular measurement system. As long as the measurement system has variable windowing I can see enough detail in the high frequency response and be confident it is looking at the near-anechoic response of the loudspeaker in the HF region.

The rest of the devil is in the details :).
 

·
Premium Member
Joined
·
129 Posts
Discussion Starter #10 (Edited)
Bobkatz,

Hi, my name is Nick. I very often read this and other forums but rarely do I chime in on a topic.windowing!

Hi, Nick. For your car measurements, windowing out the phasey issues with near walls are very important. But in the car case it's as much for practical reasons as for psychoacoustic reasons since the walls are so close they produce obvious anomalies when the window is too wide. But for psychoacoustic reasons you need a long window to assess how the ear responds to the bass.

I have been using combinations of Acourate, Room EQ Wizard, Spectrafoo and FuzzMeasure for a long time. The beauty and detail of the graphics, the ergonomics as well as the features are all important to me so I can't live without any of them. Of those four, only Acourate has a variable (psychoacoustic) window and is the most accurate of all. And that's what I use when I need to be as precise as possible. But ergonomics are not Acourate's strong point. As you can see I'm on a campaign to lobby the rest of these fine applications to implement a variable window :).

Bob
 

·
Premium Member
Joined
·
6,711 Posts
Hi Bob, if you can send me an example impulse response and a screenshot of how you prefer it to be presented I'll look at some windowing options.
 

·
Premium Member
Joined
·
129 Posts
Discussion Starter #13
Hi Bob, if you can send me an example impulse response and a screenshot of how you prefer it to be presented I'll look at some windowing options.
Thanks very much, John. Would a sample response of a real world loudspeaker be ok?

Or, what would a Dirac pulse tell you? ... but that would just look ruler flat. Not sure what you would learn I suppose unless we created some special test signals? Give me some hints as to how we would determine that other than to just do it.

If the job of a sliding window seems difficult to you, what about the idea of two or more frequency responses spliced together, one made with one window and one made with another. I know that John Atkinson frequently splices together two frequency responses, one nearfield near the ports of a loudspeaker and one farfield... so that's an approach.

Best wishes,


Bob
 

·
Premium Member
Joined
·
6,711 Posts
A speaker measurement would be preferable. There is no particular difficulty in the basic implementation, I did it back when deciding what kind of variable option to offer (as discussed in this post), but some thought would be needed around how to make the option available and how to allow the settings to be adjusted.
 

·
Premium Member
Joined
·
129 Posts
Discussion Starter #15
A speaker measurement would be preferable. There is no particular difficulty in the basic implementation, I did it back when deciding what kind of variable option to offer (as discussed in this post), but some thought would be needed around how to make the option available and how to allow the settings to be adjusted.
Dear John: I'm so glad you are willing to consider this variable feature as it is different from smoothing or averaging.

I can show you a frequency response measurement from Acourate and I could send you the impulse response of those speakers that produced that measured frequency response. Would that help?
 

·
Premium Member
Joined
·
129 Posts
Discussion Starter #16
In terms of how to specify the variable window, one method is to define the number of cycles (Hertz) that it's wide. For example, 15 cycles at 1 kHz translates to 15 ms. And so on.

I am not 100% convinced that approach gives a long enough window for psychoacoustic estimates at low frequencies when it gives a short enough window at high frequencies. However, to my ears, in my room, I have never had to manually touch or correct Acourate's correction filters so they must be doing something right, at least at the low end. The amount of treatment in the room and the room decay time surely must enter in here. My room is well treated so there isn't much to perceive after 150 milliseconds at 100 Hz.

Maybe I can ask JJ and he'll help. But he's constrained by some old NDAs so he may not have an answer. Won't hurt to ask.
 

·
Premium Member
Joined
·
6,711 Posts
I can show you a frequency response measurement from Acourate and I could send you the impulse response of those speakers that produced that measured frequency response. Would that help?
Yes, that would be great, along with a note on whatever settings you used to produce the response.
 

·
Premium Member
Joined
·
129 Posts
Discussion Starter #18 (Edited)
OK, here we go. It appears that the "default" psychoacoustic smoothing in Acourate works this way: The window width is a linear interpolation from a low frequency of 16 Hz up to Nyquist. Not sure why Nyquist, but if the Sample rate (most commonly used) is 48 kHz that would be 24 kHz, which isn't too bad as a standard.

He takes this window measure on 1/24th octave intervals.

The apparent "default" window width is specified as 15/15, which means 15 cycles at the low frequency and 15 cycles at the high frequency. These values can be changed, but I've found they produce a result that correlates well with the ear's interpretation of the response. 15 cycles would be 15 ms. at 1 kHz. To the best of what I can determine this is actually the right hand window width, and he uses a Blackman, but some experimentation is in order. Anyway, changing the left hand window size, as long as it is reasonable, does not seem to affect the displayed frequency response.

To compare REW and Acourate I took a screenshot of Acourate's psychoacoustic response of my front left speaker taken at 9 feet distance. I exported the impulse response as a 24 bit/48 khz wav (attached). I then used my usual 500 ms. Tukey with 1/6 octave smoothing to display it in REW. I then saved this as a PNG. I then brought both of these into photoshop, carefully matched the scales, extracted the Acourate curve with the magic wand tool, and overlaid them matching amplitude at 1 kHz. That's the image I've attached.

I'll give you my thoughts about these differences in my next post in this thread.
 

Attachments

·
Premium Member
Joined
·
129 Posts
Discussion Starter #19
Reactions to the frequency response differences: Well, now that I have the scales perfectly matched in Photoshop, the differences are relatively small, but I think the devil is in the details. I think the most important difference is at the high frequencies, because the red curve is near anechoic and better reflects the perceived response of the loudspeaker. Certainly the slope of the Hf curve is dramatically different. Demonstrating that "one window width does not fit all purposes". At the low frequencies, 500 ms. Tukey + 1/6 octave smoothing is remarkably close to Acourate. At mid frequencies there are differences and depending on how many EQ bands you want to generate, this may or may not be significant.

I can't swear whether the small differences between Acourate at the low frequencies are meaningful or even perceptible, but at least they are visible! In some cases we see 2 dB difference, which could be audible to a keen ear, but don't ask me to take a blind test on this! So let me summarize by saying that probably 1/6 octave smoothing with 500 ms Tukey does a very good job at low frequencies, but not at mid or highhigh. Only by having the ability to correct a system one way or another and observe the results can we reach a valid conclusion.
 

·
Premium Member
Joined
·
129 Posts
Discussion Starter #20
I think the shape of the HF in Acourate from 1 kHz up is probably right on. I say this because the Revel Gems measured in Acourate very closely follows a linear slope from 1 kHz to 20 kHz, which is I think the intent of the loudspeaker. While in the REW result it looks like a complex curve. Acourate's measurement shows that the Revel Gems also slope very linearly and naturally from 1k to 20k.

In Acourate Convolver I can easily implement my preference, a simple linear sloping target which is just about a 1/2 dB lower than the native response of the Revels. Acourate Convolver performs that and subtly smooths out the HF response variations and after correction, the loudspeakers sound very smooth, subjectively flat.

Attached is the Left front response, psychoacoustic measure, before and after correction, 1 dB/vertical div.
 

Attachments

1 - 20 of 116 Posts
Top